The materials’ degradations caused by corrosion is a major issue in industry. Their experimental study in the laboratory, necessary in most cases, often proves difficult to perform. It also has its limits, because the processes involved generally take place over long periods of time and in complex environments, which are therefore difficult to reproduce. In this context, modeling is a powerful and complementary approach to the experimental approach, insofar as it is likely to lead to the development of predictive numerical tools and/or interpretation aids.
Modeling by the cellular automata (CA) method, proposed in this thesis, is used in fields as varied as physics, biology, chemistry and social sciences.
It consists of paving a space with a network of identical cells, each being characterized at time t by a state (which is part of a predefined set of possible states) whose temporal evolution is calculated by means of rules of transition which take into account the states of neighboring cells. Its main asset is to explore the spatio-temporal dynamics of simplified representations of systems likely to be very complex in reality.
Significant advances in corrosion modeling using the CA method have been made over the past ten years at CEA/DPC/SCCME/LECA. 3D extension of existing 2D models has in particular been successfully achieved, as well as the coupling of spatially separated anodic and cathodic reactions. This made it possible to study with the same model the competition between generalized corrosion and different types of localized corrosion. 3D models of intergranular corrosion have also been developed.
In the thesis proposed here, it will be a question of developing a CA model allowing the study of corrosion processes in which the diffusion of corrosive species in solution and/or a temperature that is both variable in time and inhomogeneous in space may prove to be dimensioning (pitting and crevice corrosion, evolution of macroscopic defects). We will take advantage of two main features: firstly the equations governing diffusive transport and heat transfers are similar (they will be simulated using 3D random walks), secondly the AC method is particularly suitable for the study of phenomena involving time-dependent interfaces/boundaries.
The model developed will be implemented in C language and CUDA, in order to perform simulations on mixed CPU/GPU computers (parallel programming on graphics cards). Code development will therefore be the main activity, with simulations being performed on dedicated CEA and ENSCP machines. In order to validate the results provided by the model, reference will be made to experimental results selected from the literature and from SCCME/LECA data.